Deciphering the best intentions: Genetic and model-based mapping of individual 
variability in social inference

Andreea O. Diaconescu
Translational Neuromodeling Unit (TNU), Institute for Biomedical Engineering, 
University of Zurich/ETH Zurich

The process of how we represent othersŐ (potentially varying) intentions is a 
fundamental problem during most social transactions. This inferential process becomes 
even more crucial when we need to rely on other peopleŐs advice regarding a course of 
action. In this talk, I will introduce a meta-Bayesian modeling framework for quantifying 
individual differences in inferring othersŐ intentions during social interactions. Using 
this modeling approach, I will show that social prediction error (PE) signals are encoded 
in reward processing regions, including the dorsal striatum and the anterior insula, in 
individuals who lack a stable model of the adviser and learn about advice validity by 
striving to maximize rewards. Those with a consistent model of the adviserŐs intentions, 
on the other hand, represent social PEs in brain regions known to be associated with 
theory of mind processes, including the anterior temporal-parietal junction and the 
bilateral dorsomedial prefrontal cortex (PFC). Furthermore, I will show evidence that 
individuals who exhibit a Val-to-Met polymorphism of the Catechol O-methyl transferase 
(COMT) gene show a distinct neural representation of signed social PEs in the 
dorsolateral and dorsomedial PFC, suggesting a critical role of dopamine in learning 
about othersŐ intentions. This model-based approach for assessing individual differences 
in the neural representation of intentionality inference will be extended to studies of 
maladaptive social learning in psychiatric disorders, including schizophrenia, autism, 
and psychopathy.